Zero-Knowledge AI Inference with High Precision (ACM CCS 2025)
-
Updated
Jan 22, 2026 - C
Zero-Knowledge AI Inference with High Precision (ACM CCS 2025)
fp32-accurate Transformer layer on fp16 tensor cores: hi/lo operand splitting (fp16x3) with fused Triton kernels, 2.83x median over the fp32 reference on a free T4 at ~1e-6 error, and the 100,000-token shape running in 14 GB. Built for TikTok TechJam 2026 Track 3.
ouyezi(欧冶子)— 通用代码性能优化方法论 skill:热点驱动的基线→剖析→优化→复测→精度核查流程,跨平台、架构无关(提炼自鲲鹏技术文档,已脱厂商细节)
Posit-Enhanced Docker-based Federated Learning: Bridging Numerical Precision and Cross-Architecture Deployment in IoT Systems
A Java console project built to reinforce core concepts from COMS 2270, including classes, objects, arrays, methods, encapsulation, and basic testing.
Package floater is not the shit in the toilet. Utils for math/big.
To associate your repository with the numerical-precision topic, visit your repo's landing page and select "manage topics."